This project, modernizing and extending a system called ATLAS (Automatically Tuned Linear Algebra System) which is included in nearly every operating system and used widely for tuning linear algebra systems, aims at developing tasks and new research related to automatic tuning of high-performance computation. The work outlines a spectrum of activities, ranging from fighting bit-rot on the somewhat stale source code to developing new optimization techniques for algorithms that use non-dense representations. The infrastructure relies on various techniques to achieve accurate timing and heuristic for automatic search. The ATLAS system has been widely adopted by both industry and academia, because of the development and careful application of these ideas and because linear problems form the foundation of scientific computing. Without ATLAS, many extremely important computations simply would not be feasible. This project includes maintenance efforts that should increase the systems ability to be increasingly self-supporting. The PI will perform prototyping of ideas that lay the groundwork for later package advancement, keeping the project active, so that it remains a usable research infrastructure.
Broader Impact: The system is already widely used serving a broad community. Any improvements would result in benefit of the research and industrial community. Moreover, UTSA is a minority-serving university that provides appropriate training to underrepresented students at all levels.
